Improving multipath routing of TCP flows by network exploration

Lecture Notes in Computer Science, 2011

The IETF is developing a new transport layer solution, Mul-tiPath TCP (MPTCP), which allows to efficiently exploit several Internet paths between a pair of hosts, while presenting a single TCP connection to the application layer. From an implementation viewpoint, multiplexing flows at the transport layer raises several challenges. We first explain how this major TCP extension affects the Linux TCP/IP stack when considering the establishment of TCP connections and the transmission and reception of data over multiple paths. Then, based on our implementation of MultiPath TCP in the Linux kernel, we explain how such an implementation can be optimized to achieve high performance and report measurements showing the performance of receive buffer tuning and coupled congestion control.

Multi-path TCP (MP-TCP) has the prospective to significantly advance applications performance by using multiple paths evidently. Multipath TCP was intended and employed as a backward compatible replacement for TCP. For this reason, it exposes the standard socket API to the applications that cannot control the utilization of the different paths. This is a key feature for applications that are unaware of the multipath nature of the network. On the contrary, this is a limitation for applications that could benefit from specific knowledge to use multiple paths in a way that fits their needs. Therefore, hosts are often connected by multiple paths, but TCP restricts communications to a single path per transport connection. Resource usage within the network would be more efficient where these multiple paths able to be used concurrently. This should enhance user experience through improved resilience to network failure and higher throughput. In this paper, we have focused on MPTCP and discussed the performance issues and its solution. We believe our concept will be useful for future works of MPTCP performance evaluation

2019 IEEE 40th Sarnoff Symposium

Multipath TCP (MPTCP) can improve overall throughput of an end-to-end connection by leveraging different network paths. However, the heterogeneity of these paths can significantly hamper MPTCP's performance. In this paper, we propose to send acknowledgments (ACKs) along the lowestlatency path. This can help improve the performance of MPTCP when subflow throughput is constrained by packet loss by reacting to loss events faster. An active probing module is also developed to dynamically select the lowest-latency path against the potential change of path condition. Experiments demonstrate that overall throughput improvement generally ranges from 10% to 50%.

This paper focuses on evaluating the use of MPTCP to forward subflows in OpenFlow networks. MPTCP is a network protocol designed to forward subflows through disjointed paths. Modern networks commonly use Equal-Cost Multipath Protocol (ECMP) to split flows through distinct paths. However, even with ECMP enabled, subflows may be forwarded through the same path. MPTCP improves the multipath routing by setting subflows to be forwarded through distinct paths. As a consequence, the amount of subflows must be considered to evaluate the network throughput. In this paper, we design Multiflow to use MPTCP in OpenFlow networks. Our proposal is to improve the throughput in shared bottlenecks by forwarding subflows from a same MPTCP connection through multiple paths. We validate our approach in a testbed where shared bottlenecks occur in the link at the endpoints. The Multiflow improvement of the network performance is evaluated in experiments about resilience and end-to-end throughput. - The 29th IEEE International Conference on Advanced Information Networking and Applications (AINA-2015) Gwangju, Korea, March 25-27, 2015

Transmission control protocol performance varies considerably, depending on network and path conditions. In this paper, we discuss path conditions that affect TCP performance, from round trip delays to path capacity and buffering. We characterize throughput performance of popular TCP congestion avoidance mechanism as well as recently proposed TCP variants via open source based network experiments. We show that superior TCP performance may be achieved via careful selection of congestion avoidance mechanism, as well as parameter tuning.

2014 IEEE 15th International Conference on High Performance Switching and Routing (HPSR), 2014

Performance-optimized datacenter networks aim to handle more efficiently the growing East-West intra-cluster traffic of BigData applications. The demanding latency constraints and traffic patterns of these applications expose the inherent bottlenecks of the often oversubscribed datacenter network topologies, favoring in stead the full-bisectional bandwidth fat-trees. And yet their topological benefits may remain unrealized in practical deployments, if such fabrics use single path or flow-level (ECMP hashing) multipath routing. Here we model in detail on Layer 2 the routing performance of modern fat-tree networks using stochastic permutations of bursty traffic. We first analytically simplify and then validate by accurate simulation models that the throughputs for 'static' d-mod-k and for ECMP-like multipath routing are 63% and 47%, respectively. We also find that ECMP routing results in a wide spread of link loads under random permutation traffic, which manifests as a 3x throughput reduction for 30% of the flows. Furthermore, ECMP can lead to collisions of mouse and elephant flows, often increasing the flow completion time (FCT) of delay-sensitive flows by a factor of 10. In contrast, packet-based multipath outperforms all the others in this study. .

MPTCP is proposed by IETF working group, it allows a single TCP stream to be split across multiple paths. It has obvious benefits in performance and reliability. MPTCP has implemented in Linux-based distributions that can be compiled and installed to be used for both real and experimental scenarios. In this article, we provide performance analyses for MPTCP with a laptop connected to WiFi access point and 3G cellular network at the same time. We prove experimentally that MPTCP outperforms regular TCP for WiFi or 3G interfaces. We also compare four types of congestion control algorithms for MPTCP that are also implemented in the Linux Kernel. Results show that Alias Linked Increase Congestion Control algorithm outperforms the others in the normal traffic load while Balanced Linked Adaptation algorithm outperforms the rest when the paths are shared with heavy traffic, which is not supported by MPTCP.

Proceedings of the 2016 workshop on Applied Networking Research Workshop - ANRW 16, 2016

Multipath TCP has been recently introduced in order to allow a better resource consumption and user quality-ofexperience. This is achieved by allowing a connection between two hosts through multiple subflows. However, with the rise of middleboxes and inherent Internet ossification, the large-scale deployment of this TCP extension is difficult. In particular, a load balancer at the entry point of a data center may forward subflows to different servers, cancelling so the advantages of Multipath TCP.

37th Annual IEEE Conference on Local Computer Networks, 2012

This paper describes Flow-Path, an AllPath flowbased switching protocol that features improved load adaptive properties. Upon arrival of a new flow to the network, it explores every possible path reaching from source to destination host and selects the lowest latency path at the moment. It is based on the same basic principle than ARP-Path, that is, snooping the ARP protocol dialog (request and reply messages) to explore all available paths at the same time that address resolution takes place, but it is flow-based instead of source address-based. While preserving the main advantages of ARP-Path: shortest path bridging exploiting the full network topology, Flow-Path has the advantages of full independence of flows at the time of path creation and guarantees path symmetry (congruency) and increased path diversity. Flow-Path thus improves load distribution, at the expense of increased address table size in each bridge.

Computer Networks, 2015

IP networks were designed to provide general connectivity. At their advent, routing methods focused only on finding one optimal path between given endpoints. Although many solutions to sending traffic via multiple paths have appeared over time, the majority of current IP networks are still managed to support only single-path transmissions. This survey examines various approaches which can provide multipath transmissions in existing IP networks. Firstly, the most recognizable solutions are presented, and later, less well-known proposals are introduced. We show how it is possible to realize multipath transmission in source and hop-by-hop routing, multi-topology routing, bio-inspired routing solutions, Valiant's routing, Multi-Protocol Label Switching, Software-Defined Networks, Flow-Aware Multi-Topology Adaptive Routing, Shortest-Path Bridging, Transparent Interconnection of Lots of Links, network virtualization, and Multipath TCP. Moreover, the mentioned approaches are compared, contrasted and subjectively assessed. The goal of the survey is to show that multipath transmissions can be achieved in the current IP networks and in many different ways.